Automatic Mapping Clinical Notes to Medical - RMIT University

ous accounts of the effects of coherence relations
on pronoun resolution have taken the view that
the kind of relation shifts attention to different aspects
of the event being described (Stevenson et
al., 1994; Stevenson et al., 2000). If an event has,
for example, a start-state and an end-state, then
a narrative relation will shift attention toward the
start-state while a causal relation will shift attention
toward the end-state. Subsequent pronominal
reference will therefore prefer referents associated
with these respective states, as illustrated in (5)-
(6). Based on this argumentation, the prediction
would be that pronominal reference might favor
one non-salient referent over another if it is associated
with that part of the event to which attention
has been shifted by the coherence relation.
3 Experiment
Before describing the experiment, I’ll review the
primary and secondary questions which this experiment
is designed to test. First, there is the
question of what happens during pronoun resolution
processes when there are competing nonsalient
referents. Answers to this question should
provide evidence toward either a categorical or a
gradient model of salience ranking. Furthermore,
because investigating this question requires controlling
for syntactic versus semantic prominence
as well as coherence relation effects, two other
secondary questions are also investigated. First,
which is a more important factor in pronoun resolution:
syntactic or semantic prominence? Second,
what effect do coherence relations have on
pronominal reference to non-salient entities?
3.1 Design
The research questions described above were investigated
in this study using the well-known
spray/load-constructions which exhibit the locative
alternation (Levin, 1993) as shown in (7) and
have synonymous alternative forms. 3
(7) a. John sprayed some paint on a wall.
b. John sprayed a wall with some paint.
3 There is some difference of opinion on whether the two
forms of spray/load-constructions are actually synonymous.
One central point of contention is whether the totality effects
on the direct object (i.e., the judgment that the entity in direct
object position is totally used up in the event) are consistent
across both forms. In the judgment of Rappaport and Levin
(1988), the totality effect applies only with the with-variant.
In contrast, it is my judgment (Rose, 2005) and also that of
Tenny (1994, see her data items (100) and (102)) that the effect
applies across both forms.
92
According to prominence hierarchies in which
the syntactic subject or the semantic agent is most
prominent, then JOHN should consistently be regarded
as the (most) salient referent while PAINT
and WALL should be regarded as less or nonsalient
referents in these sentences. Thus, subsequent
pronominal reference with the third-person
singular pronoun, it, allows a test of the three different
questions outlined above.
First, if a categorical approach to salience is sufficient,
then there should be no overall preference
for either PAINT or WALL. But if gradient salience
is necessary for ranking, then it might be possible
to observe a difference between the two.
The nature of this difference, however, might
be more complex depending on the way salience
ranking is determined. If syntactic prominence is
the only relevant factor, then preferences should
consistently favor the object (i.e, PAINT in (7a),
WALL in (7b)) according to the well-established
syntactic prominence hierarchy in (2) above. But
if semantic prominence is the only factor, then
preferences should favor either the theme (PAINT)
or the location (WALL) depending on how the semantic
prominence hierarchy is ordered. One prediction
might be based on proposed thematic hierarchies
(cf., Larson (1988), Speas (1990)) which
place theme above location. According to such a
hierarchy, PAINT should be consistently preferred.
This is what I observed in Rose (2005).
Other differences may result from the kind of
coherence relation used. However, for spray/loadconstructions,
this is a little difficult to predict.
The two non-salient entities are both arguably a
part of the end-state of the event—that is, together,
they are the product of the agent’s work. Thus, any
motivation to distinguish between the two with respect
to the coherence relation must come from
some other feature of the event or its participants.
I will address the possibility in the discussion section
below.
3.2 Method
3.2.1 Participants
The participants in this experiment included 36
undergraduate students at Morehead State University
in Kentucky. Students were recruited through
fliers and classroom announcements and received
five dollars for their participation.